In a previous article, I wrote about the basics of using the new version and new feature "Style Reference" in Midjourney. (If you haven't seen it, you can see here→ 3 minutes to teach you to stably output Chinese-style illustrations, and you will be taught to do so!)
We all know that "Style Reference" can make it easy for us to get a lot of style beauty pictures we want, and the style output is very stable. So it's really addictive as soon as you use it. So today, we're going to share 3 more advanced tips. It can make us more handy!
In short, let's take a look, I hope it can help you!
01
What is a style reference
"Style reference" is actually easy to understand, which is to refer to the style of the image we provide, combined with the text prompt we set ourselves, to generate a new image.
Directive --sref
Usage: --sref URL (enter the link to the image after the command, note that there are spaces before and after, which can be one or more images, separated by spaces)
Available for Midjourney V6 and Niji V6 versions
Here are the detailed steps. If you already know it, you can go straight to the next part.
- Click the plus button in the bottom left corner, upload the image in the style you like and press enter to send it to the server.
- Right-click and select Copy Link
- Then type --sref after your keyword (note the spaces before and after it)
- Finally, paste the link to the image you copied in the previous step
02
Advanced details
Sometimes, we find that we still don't get the image we want after using the "Style Reference" function. At this point, we need some way to fine-tune the image in order to get the image we want more accurately.
- Adjust text prompts
- 1. Replace unwanted or inappropriate descriptions
- 2. Add descriptors that you need to emphasize Even if they are repetitive, including style, form, content, etc
- Interesting Cases:
- Let's say your prompt contains "a sleeping cat on the windowsill", and then the rendering is done to get a cat with open eyes. You can try to emphasize the details by repeating the description with synonyms for the keyword. Or you can even just repeat a keyword.
- 1) A cat sleeping, dozing, napping, and closing its eyes on the windowsill.
- 2) A sleeping, sleeping, sleeping cat on the windowsill. (How can there be a sense of déjà vu that important things are said three times, but there is this case in the official document, which is indeed one of the feasible methods of official certification)
- Adjust the weight of the style reference
- Directive: --sref URL --sw 100 (max 1000)
- We can adjust the weight of the style reference by adding the --sw 20 command to the end of the style reference directive, the default value is 100, that is, when no adjustment is made. Lowering the value is weakening the style, increasing the value is strengthening the style, and the maximum value is 1000.
- With this adjustment, we can control the degree of stylization more precisely. The two images below are different effects for the same style map I experimented with, with different weight values.
--sw 20
--sw 800
- Cushion drawings and style references are used in combination
- 指令： ImagePrompt url text prompt --sref url
- Content: Pad map link Text tip --sref style reference image link
- The previous image link is the earliest pad map method we used, because the previous one takes the content. The following style reference image links are taken from the style.
- In this way, we can get an image that combines content and style very directly.
- It should be noted that this method must add text prompt content in the middle, otherwise it will not be generated.
- The following 4 pictures are the last two pictures I generated with the first two pictures, and you can intuitively feel the effect.
The content of the pad diagram
The style is based on the style taken from the diagram
Generate Figure 1
Generate Figure 2
- Character references and style references are used in combination
- 指令： text prompt --cref urlA --sref urlB
- Contents: Text prompt --cref character reference image A --sref style reference image B
- The usage steps of character reference and style reference are similar, so I won't repeat the description, but the command name is different, which is --cref.
- The characters in the character reference instructions can be combined with different styles of scene images, and we can get different styles of images of the same character, or unified style images of different characters, in short, you can produce a lot of interesting combinations, you can try more.
- The following example is the third image I generated with the first two, so you can get a feel for it.
Figure A
Style reference Figure B
Generate a diagram